MOLECULAR GENETIC DETERMINANTS OF ANTIMICROBIAL RESISTANCE IN EMERGING MULTIDRUG-RESISTANT BACTERIAL PATHOGENS

Abstract

The escalating crisis of multidrug-resistant (MDR) ESKAPE pathogens is driven by complex genetic and regulatory networks that evade prediction by traditional additive models. Here, we present a comprehensive multi-omic and machine-learning framework to decode the epistatic architecture of pan-resistance in 342 clinical and environmental isolates of CRE, CRAB, and MDR P. aeruginosa. Using gap-free hybrid sequencing and targeted plasmidomics, we mapped a highly dynamic resistome characterized by novel chimeric plasmids. Spatiotemporal transcriptomics and proteomics revealed critical post-transcriptional buffering mechanisms that mitigate the fitness costs of resistance. To capture non-linear genetic interactions, we developed a Graph Neural Network (GNN) that significantly outperformed additive models (AUROC = 0.94) in predicting phenotypic resistance. Explainable AI (SHAP) identified critical epistatic drivers, including a synergistic ompK36 duplication and promoter SNP conferring a 16-fold MIC increase, alongside compensatory networks restoring bacterial fitness. CRISPR-Cas9 engineering rigorously validated these in silico predictions in vitro. By transitioning from reductionist gene catalogs to systems-level network models, this study provides an actionable blueprint for precision genomic diagnostics and the discovery of resistance-breaker therapeutics.